flowbook
flowbook — Flowchart Documentation Generator
Analyze codebase logic → setup flowbook → generate .flow.md files → verify → build.
Execute ALL phases in order. Do NOT skip phases. Generate flows for ALL significant logic — not just a few.
Phase 1: Project Analysis
1.1 Check Flowbook Existence
Check if flowbook is already set up:
package.jsonhas"flowbook"script → Already initialized, skip Phase 2- Otherwise → Proceed to Phase 2
1.2 Detect Package Manager
| Lockfile | PM |
|---|---|
bun.lock or bun.lockb |
bun |
pnpm-lock.yaml |
pnpm |
yarn.lock |
yarn |
package-lock.json |
npm |
1.3 Detect Framework & Language
Read package.json dependencies:
| Dependency | Framework |
|---|---|
next |
Next.js |
nuxt |
Nuxt |
@sveltejs/kit |
SvelteKit |
svelte (no kit) |
Svelte |
vue (no nuxt) |
Vue |
@angular/core |
Angular |
express / fastify / hono / koa |
Node.js Backend |
@nestjs/core |
NestJS |
react (no next) |
React |
django / flask / fastapi |
Python Backend |
spring |
Java/Kotlin Backend |
gin / echo / fiber |
Go Backend |
Language detection:
tsconfig.json→ TypeScript*.gofiles → Go*.pyfiles → Python*.java/*.ktfiles → Java/Kotlin- Otherwise → JavaScript
1.4 Detect Source Structure
Scan for actual source directories:
Frontend:
src/components/,src/pages/,src/views/,src/routes/src/app/,app/,pages/,components/src/store/,src/hooks/,src/composables/,src/lib/
Backend:
src/routes/,src/api/,src/controllers/,src/services/src/middleware/,src/handlers/,src/resolvers/routes/,api/,controllers/,services/
Shared:
src/utils/,src/helpers/,src/lib/src/models/,src/schemas/,src/types/
Only include directories that actually exist.
Phase 2: Flowbook Setup
2.1 Initialize Flowbook
npx flowbook@latest init
This will:
- Install
flowbookas a devDependency - Add
"flowbook"and"build-flowbook"scripts topackage.json - Create
flows/example.flow.mdas a starter template - Add
flowbook-staticto.gitignore
2.2 Verify Setup
Check that:
package.jsoncontains"flowbook": "flowbook dev"scriptflowbookis indevDependenciesflows/directory exists
2.3 Remove Example Flow
After verification, delete the example to replace with real flows:
rm flows/example.flow.md
Phase 3: Codebase Analysis & Flow Discovery
This is the most critical phase. Deeply analyze the codebase to identify all significant logic flows.
3.1 Flow Categories to Discover
Scan the codebase for these flow types. For EACH one found, plan a .flow.md file:
A. API / Route Flows
- HTTP request → middleware chain → handler → response
- REST endpoints (GET, POST, PUT, DELETE)
- GraphQL resolvers (Query, Mutation)
- WebSocket message flows
- RPC handlers
B. Authentication & Authorization
- Login / signup / logout flows
- Token refresh / session management
- OAuth flows (redirect → callback → token exchange)
- Role-based access control (RBAC) decision trees
- Password reset / email verification
C. Data Flows
- CRUD operations (Create → Validate → Save → Respond)
- Data pipeline / ETL (Extract → Transform → Load)
- Form submission → validation → API call → state update
- File upload → process → store → respond
- Cache strategies (read-through, write-through, invalidation)
D. State Management
- Global state flow (Redux, Zustand, Pinia, Vuex)
- Action → Reducer → State → UI update cycle
- Side effects (Sagas, Thunks, Effects)
- Optimistic updates → rollback on failure
E. Business Logic
- Order processing / checkout flow
- Payment flow (initiate → process → confirm / refund)
- Notification system (trigger → queue → send → track)
- Scheduling / cron job flows
- Approval workflows (submit → review → approve/reject)
F. Error Handling
- Global error boundary flow
- Retry strategies (exponential backoff)
- Fallback / circuit breaker patterns
- Error logging / monitoring pipeline
G. DevOps & Infrastructure
- CI/CD pipeline stages
- Deployment flow
- Health check / monitoring flow
- Database migration flow
H. Lifecycle & Initialization
- App bootstrap / initialization sequence
- Component lifecycle flows
- Server startup → middleware registration → route binding → listen
- Database connection → migration → seeding → ready
3.2 How to Analyze
For each source file:
- Read the file — understand its purpose
- Trace the flow — follow function calls, conditionals, async operations
- Identify decision points — if/else, switch, try/catch, early returns
- Map dependencies — what other modules/services does it call?
- Note error paths — what happens when things fail?
3.3 Flow Classification
For each discovered flow, determine:
| Field | How to Determine |
|---|---|
title |
Clear, descriptive name (e.g., "User Login Flow") |
category |
Group by domain: Authentication, API, Data, State, Business, DevOps, etc. |
tags |
Relevant keywords for filtering |
order |
Lower = more important. Core flows first. |
description |
One-line summary of what the flow does |
3.4 Skip Rules
Do NOT create flows for:
- Trivial utility functions (formatDate, slugify, etc.)
- Simple getters/setters with no logic
- Type definitions / interfaces only
- Test files
- Config files (unless they represent a complex pipeline)
- Files that already have corresponding
.flow.md
Phase 4: Flow File Generation
4.1 File Placement
Place ALL flow files in the flows/ directory at project root:
flows/
├── auth-login.flow.md
├── auth-oauth.flow.md
├── api-user-crud.flow.md
├── data-order-processing.flow.md
├── state-cart-management.flow.md
└── devops-ci-pipeline.flow.md
Naming convention: {category}-{name}.flow.md (kebab-case)
4.2 Flow File Template
Every .flow.md file MUST follow this structure:
---
title: {Descriptive Title}
category: {Category Name}
tags: [{tag1}, {tag2}, {tag3}]
order: {number}
description: {One-line description}
---
```mermaid
flowchart TD
A[Start] --> B{Decision}
B -->|Yes| C[Action]
B -->|No| D[Other Action]
C --> E[End]
D --> E
```
4.3 Mermaid Diagram Guidelines
Node Types
flowchart TD
A[Regular Step] %% Rectangle: action/process
B{Decision Point} %% Diamond: if/else, switch
C([Start / End]) %% Stadium: entry/exit points
D[(Database)] %% Cylinder: DB operations
E[[Sub-routine]] %% Subroutine: function call
F>Event / Signal] %% Flag: async event, webhook
G{{Validation}} %% Hexagon: validation step
H[/Input/] %% Parallelogram: user input
I[\Output\] %% Reverse parallelogram: response
Label Quoting Rules (MANDATORY)
Node labels containing special characters MUST be wrapped in double quotes to prevent Mermaid parse errors.
Characters that REQUIRE quoting:
| Character | Why it breaks | Unquoted (BROKEN) | Quoted (CORRECT) |
|---|---|---|---|
() |
Conflicts with ([...]) stadium and (...) rounded shapes |
A([Agent run() Start]) |
A(["Agent run() Start"]) |
{} |
Conflicts with {...} diamond shape |
B{tokio::select!{}} |
B{"tokio::select!{}"} |
[] |
Conflicts with [...] rectangle shape |
C[arr[0] value] |
C["arr[0] value"] |
:: |
Interpreted as Mermaid class/namespace syntax | D[std::io::Error] |
D["std::io::Error"] |
# |
Interpreted as Unicode escape or comment | E[Issue #42] |
E["Issue #42"] |
& |
Interpreted as HTML entity start | F[A & B] |
F["A & B"] |
Rule: When in doubt, quote it. Quoting a label that doesn't need it causes no harm. Unquoted special characters WILL break rendering.
Examples of correct quoting by node shape:
flowchart TD
A(["fn main() entry"]) %% Stadium with parens
B["process_data(input)"] %% Rectangle with parens
C{"is_valid(x)?"} %% Diamond with parens
D[["handle_error(err)"]] %% Subroutine with parens
E{{"validate(req)"}} %% Hexagon with parens
F["Config::new()"] %% Rectangle with double colon
NEVER generate unquoted labels containing (), {}, [], ::, #, or &.
Edge Labels
flowchart TD
A{Authenticated?} -->|Yes| B[Dashboard]
A -->|No| C[Login Page]
D[API Call] -->|200 OK| E[Process Data]
D -->|4xx/5xx| F[Handle Error]
G[Submit] -->|Valid| H[Save]
G -->|Invalid| I[Show Errors]
Styling — Flow-Group Coloring
Color nodes by logical flow group, not by shape. Nodes on the same path or serving the same purpose share a color.
Step 1: Identify flow groups — Analyze the diagram and group nodes by semantic role:
- Happy path / success flow
- Error / failure / rejection path
- Validation / guard checks
- External service calls (API, third-party)
- Data storage operations (DB read/write, cache)
- User interaction points (input, redirect)
- Each distinct branch from a decision point
Step 2: Assign colors using classDef + class — One classDef per group, then apply with class:
flowchart TD
A([POST /api/login]) --> B{{"Validate Input"}}
B -->|Invalid| C[\400 Bad Request/]
B -->|Valid| D[(Find User)]
D -->|Not Found| E[\401 Unauthorized/]
D -->|Found| F{{Compare Password}}
F -->|Mismatch| E
F -->|Match| G[Generate Token]
G --> H[(Save Token)]
H --> I[\200 OK + Token/]
classDef entry fill:#6366f1,stroke:#818cf8,color:#fff
classDef validation fill:#f59e0b,stroke:#fbbf24,color:#000
classDef success fill:#10b981,stroke:#34d399,color:#fff
classDef error fill:#ef4444,stroke:#f87171,color:#fff
classDef data fill:#3b82f6,stroke:#60a5fa,color:#fff
class A entry
class B,F validation
class G,I success
class C,E error
class D,H data
Color palette (assign by flow group, not by shape):
| Group | Color | Hex | Use When |
|---|---|---|---|
| Entry/Exit | Indigo | #6366f1 |
Start/end points of the flow |
| Success path | Green | #10b981 |
Happy path, successful operations |
| Error path | Red | #ef4444 |
Failures, rejections, error responses |
| Validation | Amber | #f59e0b |
Guards, checks, decision points that validate |
| Data ops | Blue | #3b82f6 |
DB reads/writes, cache, storage |
| External | Purple | #8b5cf6 |
Third-party API calls, external services |
| User action | Cyan | #06b6d4 |
User-facing interactions, redirects |
| Processing | Slate | #64748b |
Internal processing, transformation |
Rules:
- Nodes on the same logical path MUST share the same
classDef - A decision node (diamond) gets the color of the flow group it guards (e.g., validation diamond →
validationclass) - If a node belongs to multiple paths, color it by its primary purpose
- Use
classDef+class(NOT individualstyleper node) — it's cleaner and groups are explicit - Keep to 3–5 color groups per diagram. Too many colors defeats the purpose
- Uncolored nodes use the theme default — only color nodes that benefit from grouping
Subgraphs for Complex Flows
Use subgraphs to group related steps:
flowchart TD
A([Request]) --> B
subgraph Middleware
B[Auth Check] --> C[Rate Limit] --> D[Parse Body]
end
D --> E
subgraph Handler
E[Validate Input] --> F{Valid?}
F -->|Yes| G[Process]
F -->|No| H[Return 400]
end
G --> I[(Save)]
I --> J[\200 OK/]
H --> K[\400 Error/]
4.4 Complexity Guidelines
- Simple flows (3-8 nodes): Single linear or branching flow
- Medium flows (8-15 nodes): Multiple branches, some subgraphs
- Complex flows (15-25 nodes): Multiple subgraphs, parallel paths
- Do NOT exceed 25 nodes per diagram — split into multiple flows instead
If a flow is too complex:
- Create a high-level overview flow
- Create detailed sub-flows for each section
- Reference sub-flows in the overview's description
4.5 Real-World Example
For a Next.js API route app/api/auth/login/route.ts:
---
title: User Login
category: Authentication
tags: [auth, login, jwt, api]
order: 1
description: POST /api/auth/login — validates credentials and returns JWT tokens
---
```mermaid
flowchart TD
A([POST /api/auth/login]) --> B[/Parse Request Body/]
B --> C{{"Validate Email & Password"}}
C -->|Invalid| D[\400 Bad Request/]
C -->|Valid| E[(Find User by Email)]
E -->|Not Found| F[\401 Unauthorized/]
E -->|Found| G{{Compare Password Hash}}
G -->|Mismatch| F
G -->|Match| H[Generate JWT Access Token]
H --> I[Generate Refresh Token]
I --> J[(Save Refresh Token)]
J --> K[\200 OK + Tokens/]
classDef entry fill:#6366f1,stroke:#818cf8,color:#fff
classDef validation fill:#f59e0b,stroke:#fbbf24,color:#000
classDef success fill:#10b981,stroke:#34d399,color:#fff
classDef error fill:#ef4444,stroke:#f87171,color:#fff
classDef data fill:#3b82f6,stroke:#60a5fa,color:#fff
class A entry
class B,C,G validation
class H,I,K success
class D,F error
class E,J data
```
Phase 5: Verification
5.1 Syntax Check
For each generated .flow.md file:
- Verify YAML frontmatter is valid (title, category present)
- Verify mermaid code block is properly fenced (
mermaid) - Verify mermaid syntax has no obvious errors (matched brackets, valid node IDs)
- Special Character Validation (CRITICAL): Scan ALL node labels for unquoted special characters:
()inside any node shape → MUST be quoted:A(["label()"])notA([label()]){}inside any node shape → MUST be quoted:A{"label{}"}notA{label{}}[]inside any node shape → MUST be quoted:A["label[]"]notA[label[]]::anywhere in labels → MUST be quoted:A["std::io"]notA[std::io]#anywhere in labels → MUST be quoted:A["Issue #1"]notA[Issue #1]&anywhere in labels → MUST be quoted:A["A & B"]notA[A & B]- If ANY unquoted special characters are found, fix them BEFORE proceeding to build
- Verify all node IDs are unique within each diagram
- Verify subgraph labels don't contain special characters
5.2 Build Verification
npx flowbook build 2>&1
If build fails:
- Read error output
- Fix the issue (likely malformed mermaid syntax)
- Retry until build succeeds
5.3 Visual Verification
Start dev server and verify rendering:
npx flowbook dev &
FB_PID=$!
sleep 3
If the playwright skill is available, load it and:
- Navigate to
http://localhost:6200 - Wait for Flowbook UI to load
- Check sidebar — all flow categories should appear
- Click through each flow — verify diagrams render (no error messages)
- Screenshot any failures
kill $FB_PID 2>/dev/null
5.4 Fix-and-Retry Loop
If mermaid diagrams fail to render:
- Common issue: special characters in labels — wrap in quotes:
A["Label with (parens)"] - Common issue: reserved keywords — prefix with text:
A[End Point]notA[End]alone as node content after usingEndas ID - Re-run build verification
- Repeat until all diagrams render
Phase 6: Summary Report
Print a summary:
=== Flowbook Report ===
Framework: {detected framework}
Language: {detected language}
Flows generated: {N}
Categories:
- Authentication: {N} flows
- API: {N} flows
- Data: {N} flows
- State: {N} flows
- Business Logic: {N} flows
- DevOps: {N} flows
Files created:
- flows/{filename}.flow.md — {title}
- flows/{filename}.flow.md — {title}
...
Build: ✅ / ❌
Troubleshooting
Flowbook init fails
- No package.json: Run
npm init -yfirst - Permission error: Check write permissions on project directory
Mermaid syntax errors
- Brackets: Every
[,{,(must be closed - Special characters in labels: Wrap in double quotes:
A["User's Input"] - Parentheses in labels (MOST COMMON):
A([run() Start])→ Parse error. Fix:A(["run() Start"]) - Double colons in labels:
A[std::io::Error]→ Interpreted as class syntax. Fix:A["std::io::Error"] - Curly braces in labels:
B{select!{}}→ Conflicts with diamond shape. Fix:B{"select!{}"} - Square brackets in labels:
C[arr[0]]→ Conflicts with rectangle shape. Fix:C["arr[0]"] - Hash in labels:
D[Issue #42]→ Unicode escape. Fix:D["Issue #42"] - Ampersand in labels:
E[A & B]→ HTML entity. Fix:E["A & B"] - Arrow syntax: Use
-->for solid,-.->for dotted,==>for thick - Node ID reuse: Each node ID must be unique per diagram. Reuse ID to reference same node.
- Subgraph naming: Subgraph labels cannot contain special characters
Diagrams too complex
- Split into overview + detail flows
- Use subgraphs to group related logic
- Keep each diagram under 25 nodes
- Link related flows via description references
Build fails
- Check mermaid version compatibility (flowbook uses Mermaid 11+)
- Validate YAML frontmatter (no tabs, proper indentation)
- Ensure code blocks use triple backticks with
mermaidlanguage tag